How much does a Claude Code session cost?

It depends on the models the session actually used. While you stay inside your plan limits, a session costs you nothing beyond the subscription. In extra usage, the spend is billed per token: since July 7, 2026, Fable 5 bills $10 per million input tokens and $50 per million output tokens, twice the Opus 4.8 rate. One long session on the wrong model can outrun a day of normal work, which is why the per-session number is worth watching.

How do I check what a session cost?

With tokenkarma installed, type /tokenkarma status in Claude Code. The guard answers locally and instantly: today’s Claude Code spend, this session’s profile and the money the guard saved you today. The dashboard adds the longer view: model mix, costs over time and lifetime spend.

Can I stop a session from overspending?

Yes. Type /tokenkarma 20 and this session is capped at $20 of extra usage, counted from now. At the limit the prompt is blocked before it spends; resend the same prompt within 5 minutes to pass anyway. Caps act only during extra usage and are an Expert plan feature.

What did that session actually cost?

Claude Code

> /tokenkarma status

tokenkarma status
Today: $4.20 of $15.00 (28%)
This session: normal
Saves today: 2
Prices updated: 2026-07-07

Answered locally by the guard. Zero tokens, instant.

One command answers it: today's Claude Code spend, this session's profile, what the guard saved you. And the next session does not have to be a surprise: cap it at your number, and the prompt that would cross it is blocked before it spends.

Get started Cap a session

tokenkarma. Your AI costs, controlled.

Claude Code

> /tokenkarma status

tokenkarma status
Today: $4.20 of $15.00 (28%)
This session: normal
Saves today: 2
Prices updated: 2026-07-07

Answered locally by the guard. Zero tokens, instant.

The per-session number, measured locally

The guard counts your Claude Code spend on your own machine, from your own transcripts, in dollars. /tokenkarma status is answered locally: zero tokens, instant, no model call. The web dashboard adds the longer view of the same data: costs over time, lifetime spend, and the burn while a session is running.

Since July 7, 2026, the number moves faster: Fable 5 is no longer included in Claude plans and bills usage credits on every prompt. A session you left on the flagship keeps billing flagship rates until you notice. What changed on July 7, in three lines.

Which sessions actually needed Fable 5?

The model mix is the honest answer to "was that worth flagship rates?". Most sessions are routine: refactors, tests, plumbing a cheaper model handles fine. A few genuinely need the most capable tier. Your dashboard shows your own split per model, in measured dollars, so the next "which model?" decision is based on your work, not a hunch.

See it, then switch smarter. tokenkarma never silently moves your main work to another model.

It also saves while you work

Cheaper-model tips

Get a tip when a cheaper model can do the job, with the estimated saving for that prompt. Demanding work is never downgraded, and micro-prompts stay silent. Follow a tip and the saving joins your measured total; ignore it and nothing is counted.

Auto-switch

Simple background tasks run on a cheaper model, automatically. You save without doing anything. Every dollar it credits is measured from the real tokens consumed, never estimated, and measured and estimated numbers are never added together.

Knowing the number is half of it. Capping it is the other half.

Type /tokenkarma 20 at the start of a session and it stops at $20 of extra usage, counted from now. The one-command cap is the daily gesture; the full control-by-control comparison is on the cap page. The short version:

Anthropic native

One spending cap, applied per month. Nothing per session, per day or per week. When the credits run out, Fable 5 stops mid-session.

tokenkarma

A cap on Claude Code spend per session, per day, per week or per month. The prompt is blocked before it spends, at the number you chose.

Blocked before it spends. Resend the same prompt to pass anyway. You stay in charge.

Anthropic controls as documented in Anthropic’s own support articles on extra usage and spending limits, verified July 2026.

Seeing is free: every plan gets the status readout, the per-session costs and the live burn. The caps themselves are an Expert feature; on Free and Hobbyist they show as locked. See pricing.

Frequently asked questions

How much does a Claude Code session cost?: It depends on the models the session actually used. While you stay inside your plan limits, a session costs you nothing beyond the subscription. In extra usage, the spend is billed per token: since July 7, 2026, Fable 5 bills $10 per million input tokens and $50 per million output tokens, twice the Opus 4.8 rate. One long session on the wrong model can outrun a day of normal work, which is why the per-session number is worth watching.
How do I check what a session cost?: With tokenkarma installed, type /tokenkarma status in Claude Code. The guard answers locally and instantly: today’s Claude Code spend, this session’s profile and the money the guard saved you today. The dashboard adds the longer view: model mix, costs over time and lifetime spend.
Which sessions actually needed Fable 5?: That is the question the model mix answers. Most sessions are routine work a cheaper model handles fine; a few genuinely need the flagship. Seeing your own split is what lets you pick smarter next time. tokenkarma shows it; it never silently moves your main work to another model.
Can I stop a session from overspending?: Yes. Type /tokenkarma 20 and this session is capped at $20 of extra usage, counted from now. At the limit the prompt is blocked before it spends; resend the same prompt within 5 minutes to pass anyway. Caps act only during extra usage and are an Expert plan feature.